Performance – Vectorize MCTS Selection

The PUCT formula in MCTS._simulate is implemented in a Python loop.

**Problem:**
1. Python loops are slow for performance-critical code
2. Selection bottlenecks MCTS speed

**Impact:**
1. Slower search simulations
2. Limits self-play throughput

**Recommendation:**
1. Store node statistics (Q, N, P) in NumPy arrays
2. Perform PUCT calculation for all child nodes in a single vectorized operation